Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers—8× deeper than VGG nets [40] but still having lower complexity. An ensemble of these residual nets achieves 3.57% error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers. The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28% relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions1, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.

/pdf/deep-residual-learning-for-image-recognition-b75jg61qrq.pdf

Deep Residual Learning for Image Recognition

We introduce Adam, an algorithm for first-order gradient-based optimization of stochastic objective functions, based on adaptive estimates of lower-order moments. The method is straightforward to implement, is computationally efficient, has little memory requirements, is invariant to diagonal rescaling of the gradients, and is well suited for problems that are large in terms of data and/or parameters. The method is also appropriate for non-stationary objectives and problems with very noisy and/or sparse gradients. The hyper-parameters have intuitive interpretations and typically require little tuning. Some connections to related algorithms, on which Adam was inspired, are discussed. We also analyze the theoretical convergence properties of the algorithm and provide a regret bound on the convergence rate that is comparable to the best known results under the online convex optimization framework. Empirical results demonstrate that Adam works well in practice and compares favorably to other stochastic optimization methods. Finally, we discuss AdaMax, a variant of Adam based on the infinity norm.

Adam: A Method for Stochastic Optimization

We trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 different classes. On the test data, we achieved top-1 and top-5 error rates of 37.5% and 17.0% which is considerably better than the previous state-of-the-art. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully-connected layers with a final 1000-way softmax. To make training faster, we used non-saturating neurons and a very efficient GPU implementation of the convolution operation. To reduce overriding in the fully-connected layers we employed a recently-developed regularization method called "dropout" that proved to be very effective. We also entered a variant of this model in the ILSVRC-2012 competition and achieved a winning top-5 test error rate of 15.3%, compared to 26.2% achieved by the second-best entry.

/pdf/imagenet-classification-with-deep-convolutional-neural-1190x9a6b9.pdf

ImageNet Classification with Deep Convolutional Neural Networks

Learning to store information over extended time intervals by recurrent backpropagation takes a very long time, mostly because of insufficient, decaying error backflow. We briefly review Hochreiter's (1991) analysis of this problem, then address it by introducing a novel, efficient, gradient based method called long short-term memory (LSTM). Truncating the gradient where this does not do harm, LSTM can learn to bridge minimal time lags in excess of 1000 discrete-time steps by enforcing constant error flow through constant error carousels within special units. Multiplicative gate units learn to open and close access to the constant error flow. LSTM is local in space and time; its computational complexity per time step and weight is O. 1. Our experiments with artificial data involve local, distributed, real-valued, and noisy pattern representations. In comparisons with real-time recurrent learning, back propagation through time, recurrent cascade correlation, Elman nets, and neural sequence chunking, LSTM leads to many more successful runs, and learns much faster. LSTM also solves complex, artificial long-time-lag tasks that have never been solved by previous recurrent network algorithms.

Long short-term memory

In this work we investigate the effect of the convolutional network depth on its accuracy in the large-scale image recognition setting. Our main contribution is a thorough evaluation of networks of increasing depth using an architecture with very small (3x3) convolution filters, which shows that a significant improvement on the prior-art configurations can be achieved by pushing the depth to 16-19 weight layers. These findings were the basis of our ImageNet Challenge 2014 submission, where our team secured the first and the second places in the localisation and classification tracks respectively. We also show that our representations generalise well to other datasets, where they achieve state-of-the-art results. We have made our two best-performing ConvNet models publicly available to facilitate further research on the use of deep visual representations in computer vision.

Very Deep Convolutional Networks for Large-Scale Image Recognition

We introduce linear relational embedding as a means of learning a distributed representation of concepts from data consisting of binary relations between these concepts. The key idea is to represent concepts as vectors, binary relations as matrices, and the operation of applying a relation to a concept as a matrix-vector multiplication that produces an approximation to the related concept. A representation for concepts and relations is learned by maximizing an appropriate discriminative goodness function using gradient ascent. On a task involving family relationships, learning is fast and leads to good generalization.

/pdf/learning-distributed-representations-of-concepts-using-4unif83s4r.pdf

Learning distributed representations of concepts using linear relational embedding

In this paper we develop a class of nonlinear generative models for high-dimensional time series. We first propose a model based on the restricted Boltzmann machine (RBM) that uses an undirected model with binary latent variables and real-valued "visible" variables. The latent and visible variables at each time step receive directed connections from the visible variables at the last few time-steps. This "conditional" RBM (CRBM) makes on-line inference efficient and allows us to use a simple approximate learning procedure. We demonstrate the power of our approach by synthesizing various sequences from a model trained on motion capture data and by performing on-line filling in of data lost during capture. We extend the CRBM in a way that preserves its most important computational properties and introduces multiplicative three-way interactions that allow the effective interaction weight between two variables to be modulated by the dynamic state of a third variable. We introduce a factoring of the implied three-way weight tensor to permit a more compact parameterization. The resulting model can capture diverse styles of motion with a single set of parameters, and the three-way interactions greatly improve its ability to blend motion styles or to transition smoothly among them. Videos and source code can be found at http://www.cs.nyu.edu/~gwtaylor/publications/jmlr2011.

/pdf/two-distributed-state-models-for-generating-high-dimensional-22tehf197h.pdf

Two Distributed-State Models For Generating High-Dimensional Time Series

We describe how the pretraining algorithm for Deep Boltzmann Machines (DBMs) is related to the pretraining algorithm for Deep Belief Networks and we show that under certain conditions, the pretraining procedure improves the variational lower bound of a two-hidden-layer DBM. Based on this analysis, we develop a different method of pretraining DBMs that distributes the modelling work more evenly over the hidden layers. Our results on the MNIST and NORB datasets demonstrate that the new pretraining algorithm allows us to learn better generative models.

/pdf/a-better-way-to-pretrain-deep-boltzmann-machines-2cl4zjnmgk.pdf

A Better Way to Pretrain Deep Boltzmann Machines

Phoneme recognition: neural networks vs. hidden Markov models.

Realistic facial expression animation requires a powerful “animator” (or graphics program) that can represent the kinds of variations in facial appearance that are both possible and likely to occur in a given context. If the goal is fully determined as in character animation for film, knowledge can be provided in the form of human higher-level descriptions. However, for generating facial expressions for interactive interfaces, such as animated avatars, correct expressions for a given context must be generated on the fly. A simple solution is to rely on a set of prototypical expressions or basis shapes that are linearly combined to create every facial expression in an animated sequence (Kleiser, 1989; Parke, 1972). An innovative algorithm for fitting basis shapes to images was proposed by Blanz and Vetter (1999). The main problem with the basis shape approach is that the full range of appearance variation required for convincing expressive behavior is far beyond the capacity of what a small set of basis shapes can accommodate. Moreover, even if many expression components are used to create a repertoire of basis shapes (Joshi et al., 2007; Lewis et al., 2000), the interface may need to render different identities or mixtures of facial expressions not captured by the learned basis shapes. A representation of facial appearance for animation must be powerful enough to capture the right constraints for realistic expression generation yet flexible enough to accommodate different identities and behaviors. Besides the obvious utility of such a representation to animated facial interfaces, a good model of facial expression generation would be useful for computer vision tasks because the model’s representation would likely be much richer and more informative than the original pixel data. For example, inferring the model’s representation corresponding to a given image might even allow transferring an expression extracted from an image of a face onto a different character as illustrated by the method of expression cloning (Noh & Neumann, 2001). In this chapter we introduce a novel approach to learning to generate facial expressions that uses a deep belief net (Hinton et al., 2006). The model can easily accommodate different constraints on generation. We demonstrate this by restricting it to generate expressions with a given identity and with elementary facial expressions such as “raised eyebrows.” The deep

/pdf/generating-facial-expressions-with-deep-belief-nets-59o3i3sx4v.pdf

Geoffrey E. Hinton

Papers

Learning distributed representations of concepts using linear relational embedding

Two Distributed-State Models For Generating High-Dimensional Time Series

A Better Way to Pretrain Deep Boltzmann Machines

Phoneme recognition: neural networks vs. hidden Markov models.

Generating Facial Expressions with Deep Belief Nets